Page 105 - Read Online
P. 105
Page 2 of 31 Songthumjitti et al. Intell Robot 2023;3(3):306-36 I http://dx.doi.org/10.20517/ir.2023.20
1. INTRODUCTION
Robots are created to perform tasks more efficiently and accurately than human operators. In situations that
are too dangerous or inaccessible for humans, deploying robots can provide a safer and more effective solution.
However, in order to make a robot perform tasks, instructions must be given. While simple robot movements
can be controlled with buttons, joysticks, or computer codes, tracing human movement while they perform
difficult tasks is challenging. Another solution is to use a human–robot collaboration system, which enables
direct interaction between human operators and robots [1–3] . In particular, power-assist systems and direct
robot teaching are examples of applications for such systems in which human operators and robots physically
[4]
interact . In such systems, human operators will be able to guide the movements of a robot at will. To realize
such systems, the robot must be properly controlled based on the manipulation force of the operator.
[5]
Admittance control is one method to realize such a system . A human-machine system can be a closed-loop
system consisting of an admittance-controlled robot and a human operator. Interaction forces are generated
by end-effectors that are touched by humans, measured by force sensors, and input into the admittance model.
However, admittance control has the disadvantage that the stability of the system is affected by the stiffness
of the environment, the time delay of the control, and the bandwidth of the actuator [6,7] . In human-machine
[8]
systems, the stiffness of the human and the stiffness of the structure of a robot are particularly important .
When human stiffness increases, to prevent robots from oscillating, the admittance model must be changed
[9]
by increasing the viscosity and mass parameters . Several variable admittance control methods have been
proposed to improve operability [10] , but the variable parameter range is limited to maintain stability [11–15] . If
thestructuralcharacteristicsoftheoriginalrobotsystemhavealargeinfluence,therangeofvariableparameters
will be narrow, and the improvement in operability will be limited. Especially in the case of cartesian robots,
where the mass of the movable actuator is large, the force required to move that actuator itself is greater, so the
effect of insufficient structural stiffness of the robot is greater. This requires a particularly large mass parameter
in the admittance model to ensure stability, resulting in poor maneuverability.
In this study, we focused on improving the stability of the human-machine system without decreasing the ad-
mittance characteristic of the system. The idea is to make a compensator that can reduce the effect of structural
characteristics that affect system stability to a minimum. Buerger et al. propose a method for designing an
actuator controller for an interactive robot with structural resonance [16] . This approach is similar to ours, as
it uses a feedback controller to compensate for the structural characteristics of the robot and shape the open-
loop characteristics to improve stability. However, it assumes force control of the actuator and is difficult to
incorporate into admittance control, which is position-based impedance control. Therefore, we propose a new
method to compensate for the structural characteristics of the robot that can be easily applied to admittance
control.
To design a compensator for use in the system, we have to know the characteristics of the robot, which consist
of an actuator system and structural characteristics, which can be obtained through a spectrum analysis exper-
iment [17] . After that, a compensator can be designed to achieve a wider stability range in operation. Therefore,
we can use lower inertia and viscosity parameters in the admittance model, which means the operator can
work in a lighter environment and extend the operation duration before starting to get tired. We consider
feed-forward and feedback compensators. Feed-forward compensators are simple and require no extra sen-
sors, but they are not robust to changes in the structure characteristics. In contrast, feedback compensators are
robust but require sensors to measure the motion of the structure. Motion sensors on actuators (typically ro-
tary encoders, etc.) are relatively internal sensors and cannot measure absolute motion, which is important for
interaction with the operator outside the robot. In this study, the absolute motion of the structure is measured
by accelerometers that can be easily attached to the structure of the robot.
The results from simulations show that by appending feed-forward and feedback compensation, they can ex-